The comparison of the composition and volume of mid-ocean ridge basalts to predicted values inferred from experimental and theoretical studies is one of the primary tools for mapping thermal structure and the geometry of melting beneath mid-ocean ridges. Several indicators of either the pressure or degree of melting are commonly employed: Na2O concentration in melts is roughly inversely proportional to the extent of melting; (Sm/Yb)N is affected by the proportion of melt that is generated in the garnet stability field (deeper than 60 km); and FeO content increases with increasing pressure of melting. Crustal thickness also depends on the geometry and extent of melting. K2O/TiO2 is used as an indicator of the enrichment of the mantle source. For normal ridges, we find that regional averages of Fe(8) (FeO normalized to 8 wt % MgO to correct for fractionation) are strongly dependent on enrichment. Thus if Fe(8) is used to infer to the pressure of melting, correction for the effects of compositional heterogeneity of the source is needed. Even with the correction for heterogeneity, however, it is difficult to correct for all fractionation effects accurately enough to use FeO as a depth indicator except in the most extreme cases.

Mantle heterogeneity also affects Na(8) and (Sm/Yb)N, but the correction for enrichment by linear regression on the Na(8) and (Sm/Yb)N to a common reference value of K2O/TiO2 does not change the relationship between (Sm/Yb)N and Na(8) that is used to estimate depth and degree of melting. In simple melting models, the composition and crustal thickness depend on the rate of pressure release melting per kilometer of uplift; the initial depth of onset of melting; the final depth of termination of melting; the nature of melt equilibration, i.e., batch or fractional melting; and the geometry of upwelling, i.e., passive or dynamic flow. We find that melting beneath normal ridges commences in a narrow depth range in the spinel-garnet transition zone (about 60--70 km in depth), suggesting a much smaller variation in potential temperature beneath normal ridges (<60¿K) than previously suggested (250¿K, Klein and Langmuir, 1987; Langmuir et al., 1992). The difference is due primarily to the component of Fe(8) variation caused by mantle heterogeneity in the global data set and to difficulties in correcting FeO for fractionation. The initial depth of melting beneath hotspots is greater than beneath normal ridges, indicating significantly higher temperatures. Instead of melting continuing to the base of the crust beneath normal ridges, it ceases at a variety of depths beneath the crust. Variations in final depth of melting may be due to cooling to the surface, effects of transform faults, and the local, relative importance of dynamic upwelling. ¿ American Geophysical Union 1995